Format Converters

1. Format Syntax

StreamDevice format converters work very similar to the format converters of the C functions printf() and scanf(). But StreamDevice provides more different converters and you can also write your own converters. Formats are specified in quoted strings as arguments of out or in commands.

A format converter consists of

An exception is the sequence %% which stands for a single literal %. This has been added for compatibility with the C functions printf() and scanf(). It behaves the same as the escaped percent \%.

The flags *# +0- work like in the C functions printf() and scanf(). The flags ?, = and ! are extensions.

The * flag skips data in input formats. Input is consumed and parsed, a mismatch is an error, but the read data is dropped. This is useful if input contains more than one value. Example: in "%*f%f"; reads the second floating point number.

The # flag may alter the format, depending on the converter (see below).

The ' ' (space) and + flags usually print a space or a + sign before positive numbers, where negative numbers would have a -. Some converters may redefine the meaning of these flags (see below).

The 0 flag usually says that numbers should be left padded with 0 if width is larger than required. Some converters may redefine the meaning of this flag (see below).

The - flag usually specifies that output is left justified if width is larger than required. Some converters may redefine the meaning of this flag (see below).

The ? flag makes failing input conversions succeed with a default zero value (0, 0.0, or "", depending on the format type).

The = flag allows to compare input with current values. It is only allowed in input formats. Instead of reading a new value from input, the current value is formatted (like for output) and then compared to the input.

The ! flag demands that input is exactly width bytes long (normally width defines the maximum number of bytes read in many formats). This feature has been added by Klemen Vodopivec, SNS.

Examples:

in "%f"; Read a float value
out "%(HOPR)7.4f"; Write the HOPR field as 7 char float with precision 4
out "%#010x"; Write a 0-padded 10 char hex integer using the alternative format (with leading 0x)
in "%[_a-zA-Z0-9]"; Read a string of alphanumerical chars or underscores
in "%*i"; Skip over an integer number
in "%?d"; Read a decimal number or if that fails pretend that value was 0
in "%=.3f"; Assure that the input is equal to the current value formatted as a float with precision 3
in "%!5d"; Expect exactly 5 decimal digits. Fewer digits are considered loss of data and make the format fail.
in "%d%%"; Read a decimal number followed by a % sign

2. Data Types and Record Fields

Default fields

Every conversion character corresponds to one of the data types DOUBLE, LONG, ULONG, ENUM, or STRING. In contrast to to the C functions printf() and scanf(), it is not required to specify a variable for the conversion. The variable is typically the VAL or RVAL field of the record, selected automatically depending on the data type. Not all data types make sense for all record types. Refer to the description of supported record types for details.

StreamDevice makes no difference between float and double nor between short, int and long values. Thus, data type modifiers like l or h do not exist in StreamDevice formats.

I/O Redirection to other records or fields

To use formats with other than the default fields of a record or even with fields of other records on the same IOC, use the syntax %(record.FIELD). If only a field name but no record is given, the active record is assumed. If only a record name but no field name is given, the VAL field is assumed.

Example 1: out "%(EGU)s"; outputs the EGU field of the active record.

Example 2: in "%(otherrecord.RVAL)i"; stores the received integer value in the RVAL field of the other record and then processes that record. The other record should probably use DTYP="Raw Soft Channel" in order to convert RVAL to VAL.

Example 3: in "%(otherrecord)f"; stores the received floating point value in the VAL field of the other record and then processes that record. The other record should probably use DTYP="Soft Channel". In the unlikely case that the name of the other record is the same as a field of the active record (e.g. if you name a record "DESC"), then use .VAL explicitly to refer to the record rather than the field of the active record.

This feature is quite useful in the case that one line of input contains more than one value that need to be stored in multiple records or if one line of output needs to be contructed from values of multiple records. In order to avoid using full record names in the protocol file, it is recommended to pass the name or part of the name (e.g. the device prefix) of the other record as a protocol argument. In that case the redirection usually looks like this: in "%(\$1recordpart)f" and the record calls the protocol like this: field(INP, "@protocolfile protocol($(PREFIX)) $(PORT)") using a macro for the prefix part which is then used for \$1.

If the other record is passive and the field has the PP attribute (see Record Reference Manual), the record will be processed. It is your responsibility that the data type of the record field is compatible to the the data type of the converter. STRING formats are compatible with arrays of CHAR or UCHAR.

Be aware that using this syntax is by far not as efficient as using the default field. At the moment it is not possible to set the other record to an alarm state if anything fails. It will simply not be processed if the fault happens before or while handling it and it will already have been processed if the fault happens later.

Pseudo-converters

Some formats are not actually converters. They format data which is not stored in a record field, such as a checksum or regular expression substitution. No data type corresponds to those pseudo-converters and the %(FIELD) syntax cannot be used.

3. Standard DOUBLE Converters (%f, %e, %E, %g, %G)

Output: %f prints fixed point, %e prints exponential notation and %g prints either fixed point or exponential depending on the magnitude of the value. %E and %G use E instead of e to separate the exponent.

With the # flag, output always contains a period character.

Input: All these formats are equivalent. Leading whitespaces are skipped.

With the # flag additional whitespace between sign and number is accepted.

When a maximum field width is given, leading whitespace only counts to the field witdth when the space flag is used.

4. Standard LONG and ULONG Converters (%d, %i, %u, %o, %x, %X)

Output: %d and %i print signed decimal, %u unsigned decimal, %o unsigned octal, and %x or %X unsigned hexadecimal. %X uses upper case letters.

With the # flag, octal values are prefixed with 0 and hexadecimal values with 0x or 0X.

Unlike printf, %x and %X truncate the output to the the given width (number of least significant half bytes).

Input: %d matches signed decimal, %u matches unsigned decimal, %o unsigned octal. %x and %X both match upper or lower case unsigned hexadecimal. Octal and hexadecimal values can optionally be prefixed. %i matches any integer in decimal, or prefixed octal or hexadecimal notation. Leading whitespaces are skipped.

With the - negative octal and hexadecimal values are accepted.

With the # flag additional whitespace between sign and number is accepted.

When a maximum field width is given, leading whitespace only counts to the field witdth when the space flag is used.

5. Standard STRING Converters (%s, %c)

Output: %s prints a string. If precision is specified, this is the maximum string length. %c is a LONG format in output, printing one character!

Input: %s matches a sequence of non-whitespace characters and %c matches a sequence of not-null characters. The maximum string length is given by width. The default width is infinite for %s and 1 for %c. Leading whitespaces are skipped with %s except when the space flag is used but not with %c. The empty string matches.

With the # flag %s matches a sequence of not-null characters instead of non-whitespace characters.

With the 0 flag %s pads with 0 bytes instead of spaces.

6. Standard Charset STRING Converter (%[charset])

This is an input-only format. It matches a sequence of characters from charset. If charset starts with ^, the format matches all characters not in charset. Leading whitespaces are not skipped.

Example: %[_a-z] matches a string consisting entirely of _ (underscore) or letters from a to z.

7. ENUM Converter (%{string0|string1|...})

This format maps an unsigned integer value on a set of strings. The value 0 corresponds to string0 and so on. The strings are separated by |.

Example: %{OFF|STANDBY|ON} mapps the string OFF to the value 0, STANDBY to 1 and ON to 2.

When using the # flag it is allowed to assign integer values to the strings using =. Unassigned strings increment their values by 1 as usual.

If one string is the initial substing of another, the substing must come later to ensure correct matching. In particular if one string is the emptry string, it must be the last one because it always matches. Use # and = to renumber if necessary.

Use the assignment =? for the last string to make it the default value for output formats.

Example: %#{neg=-1|stop|pos|fast=10|rewind=-10}.

If one of the strings contains | or } (or = if the # flag is used) a \ must be used to escape the character.

Output: Depending on the value, one of the strings is printed, or the default if given and no value matches.

Input: If any of the strings matches, the value is set accordingly.

8. Binary LONG or ULONG Converter (%b, %Bzo)

This format prints or scans an unsigned integer represented as a binary string (one character per bit). The %b format uses the characters 0 and 1. With the %B format, you can choose two other characters to represent zero and one. With the # flag, the bit order is changed to little endian, i.e. least significant bit first.

Examples: %B.! or %B\x00\xff. %B01 is equivalent to %b.

In output, if width is larger than the number of significant bits, then the flag 0 means that the value should be padded with with the chosen zero character instead of spaces. If precision is set, it means the number of significant bits. Otherwise, the highest 1 bit defines the number of significant bits.

In input, leading spaces are skipped. A maximum of width characters is read. Conversion stops with the first character that is not the zero or the one character.

9. Raw LONG or ULONG Converter (%r)

The raw converter does not really "convert". A signed or unsigned integer value is written or read in the internal (usually two's complement) representation of the computer. The normal byte order is big endian, i.e. most significant byte first. With the # flag, the byte order is changed to little endian, i.e. least significant byte first. With the 0 flag, the value is unsigned, otherwise signed.

In output, the precision (or sizeof(long) whatever is less) least significant bytes of the value are sign extended or zero extended (depending on the 0 flag) to width bytes. The default for precision is 1. Thus if you do not specify the precision, only the least significant byte is written! It is common error to write out "%2r"; instead of out "%.2r";.

In input, width bytes are read and put into the value. If width is larger than the size of a long, only the least significant bytes are used. If width is smaller than the size of a long, the value is sign extended or zero extended, depending on the 0 flag.

Examples: out "%.2r"; in "%02r";

10. Raw DOUBLE Converter (%R)

The raw converter does not really "convert". A float or double value is written or read in the internal (maybe IEEE) representation of the computer. The normal byte order is big endian, i.e. most significant byte first. With the # flag, the byte order is changed to little endian, i.e. least significant byte first. The width must be 4 (float) or 8 (double). The default is 4.

11. Packed BCD (Binary Coded Decimal) LONG or ULONG Converter (%D)

Packed BCD is a format where each byte contains two binary coded decimal digits (0 ... 9). Thus a BCD byte is in the range from 0x00 to 0x99. The normal byte order is big endian, i.e. most significant byte first. With the # flag, the byte order is changed to little endian, i.e. least significant byte first. The + flag defines that the value is signed, using the upper half of the most significant byte for the sign. Otherwise the value is unsigned.

In output, precision decimal digits are printed in at least width output bytes. Signed negative values have 0xF in their most significant half byte followed by the absolute value.

In input, width bytes are read. If the value is signed, a one in the most significant bit is interpreted as a negative sign. Input stops with the first byte (after the sign) that does not represent a BCD value, i.e. where either the upper or the lower half byte is larger than 9.

12. Checksum Pseudo-Converter (%<checksum>)

This is not a normal "converter", because no user data is converted. Instead, a checksum is calculated from the input or output. The width field is the byte number from which to start calculating the checksum. Default is 0, i.e. the first byte of the input or output of the current command. The last byte is precision bytes before the checksum (default 0). For example in "abcdefg%<xor>" the checksum is calculated from abcdefg, but in "abcdefg%2.1<xor>" only from cdef.

Normally, multi-byte checksums are in big endian byteorder, i.e. most significant byte first. With the # flag, the byte order is changed to little endian, i.e. least significant byte first.

The 0 flag changes the checksum representation to hexadecimal ASCII (2 chars per checksum byte).

The - flag changes the checksum representation to "poor man's hex": 0x30 ... 0x3f (2 chars per checksum byte).

The + flag changes the checksum representation to decimal ASCII (formatted with %d).

In output, the checksum is appended.

In input, the next byte or bytes must match the checksum.

Implemented checksum functions

%<sum> or %<sum8>
One byte. The sum of all characters modulo 28.
%<sum16>
Two bytes. The sum of all characters modulo 216.
%<sum32>
Four bytes. The sum of all characters modulo 232.
%<negsum>, %<nsum>, %<-sum>, %<negsum8>, %<nsum8>, or %<-sum8>
One byte. The negative of the sum of all characters modulo 28.
%<negsum16>, %<nsum16>, or %<-sum16>
Two bytes. The negative of the sum of all characters modulo 216.
%<negsum32>, %<nsum32>, or %<-sum32>
Four bytes. The negative of the sum of all characters modulo 232.
%<notsum> or %<~sum>
One byte. The bitwise inverse of the sum of all characters modulo 28.
%<xor>
One byte. All characters xor'ed.
%<xor7>
One byte. All characters xor'ed & 0x7F.
%<crc8>
One byte. An often used 8 bit crc checksum (poly=0x07, init=0x00, xorout=0x00).
%<ccitt8>
One byte. The CCITT standard 8 bit crc checksum (poly=0x31, init=0x00, xorout=0x00, reflected).
%<crc16>
Two bytes. An often used 16 bit crc checksum (poly=0x8005, init=0x0000, xorout=0x0000).
%<crc16r>
Two bytes. An often used reflected 16 bit crc checksum (poly=0x8005, init=0x0000, xorout=0x0000, reflected).
%<modbus>
Two bytes. The modbus 16 bit crc checksum (poly=0x8005, init=0xffff, xorout=0x0000, reflected)
%<ccitt16>
Two bytes. The usual (but wrong?) implementation of the CCITT standard 16 bit crc checksum (poly=0x1021, init=0xFFFF, xorout=0x0000).
%<ccitt16a>
Two bytes. The unusual (but correct?) implementation of the CCITT standard 16 bit crc checksum with augment. (poly=0x1021, init=0x1D0F, xorout=0x0000).
%<ccitt16x> or %<crc16c> or %<xmodem>
Two bytes. The XMODEM checksum. (poly=0x1021, init=0x0000, xorout=0x0000).
%<crc32>
Four bytes. The standard 32 bit crc checksum. (poly=0x04C11DB7, init=0xFFFFFFFF, xorout=0xFFFFFFFF).
%<crc32r>
Four bytes. The standard reflected 32 bit crc checksum. (poly=0x04C11DB7, init=0xFFFFFFFF, xorout=0xFFFFFFFF, reflected).
%<jamcrc>
Four bytes. Another reflected 32 bit crc checksum. (poly=0x04C11DB7, init=0xFFFFFFFF, xorout=0x00000000, reflected).
%<adler32>
Four bytes. The Adler32 checksum according to RFC 1950.
%<hexsum8>
One byte. The sum of all hex digits. (Other characters are ignored.)
%<lrc>
One byte. The Longitudinal Redundancy Check according to Wikipedia.
%<hexlrc>
One byte. The LRC for the hex digits. (Other characters are ignored.)
%<leybold>
One byte. Used by some Leybold products. 255-bytesum%255 (+32 if result would be <32)
%<brksCryo>
One byte. Used by Brooks Cryopumps.
%<CPI>
One byte. Used by TRIUMF CPI RF amplifier.
%<bitsum> or %<bitsum8>
One byte. Number of 1 bits in all characters.
%<bitsum16>
Two bytes. Number of 1 bits in all characters.
%<bitsum32>
Four bytes. Number of 1 bits in all characters.

13. Regular Expresion STRING Converter (%/regex/)

This input-only format matches Perl compatible regular expressions (PCRE). It is only available if a PCRE library is installed.

If PCRE is not available for your host or cross architecture, download the sourcecode from www.pcre.org and try my EPICS compatible Makefile to compile it like a normal EPICS support module. The Makefile is known to work with EPICS 3.14.8 and PCRE 7.2. In your RELEASE file define the variable PCRE so that it points to the install location of PCRE.

If PCRE is already installed on (some of) your systems, you may add architectures where PCRE can be found in standard include and library locations to the variable WITH_SYSTEM_PCRE. If either the header file or the library are in a non-standard place, set in your RELEASE file the variables PCRE_INCLUDE_arch and/or PCRE_LIB_arch for the respective architectures to the correct directories or set PCRE_INCLUDE and/or PCRE_LIB in architecture specific RELEASE.Common.arch files.

If the regular expression is not anchored, i.e. does not start with ^, leading non-matching input is skipped. To match in multiline mode (across newlines) add (?m) at the beginning of the pattern. To match case insensitive, add (?i).

A maximum of width bytes is matched, if specified. If precision is given, it specifies the sub-expression in () whose match is returned. Otherwise the complete match is returned. In any case, the complete match is consumed from the input buffer. If the expression contains a / it must be escaped like \/.

Example: %.1/<title>(.*)<\/title>/ returns the title of an HTML page, skipps anything before the <title> tag and leaves anything after the </title> tag in the input buffer.

14. Regular Expresion Substitution Pseudo-Converter (%#/regex/subst/)

This is a variant of the previous converter (note the #) but instead of returning the matching string, it can be used as a pre-processor for input or as a post-processor for output.

Matches of the regex are replaced by the string subst with all & in subst replaced with the match itself and all \1 through \9 replaced with the match of the corresponding sub-expression if such a sub-expression exists. Occurrences of \Un, \Ln, \un, or \ln with n being a number 0 through 9 or & are replaced with the corresponding sub-expression converted to all upper case, all lower case, first letter upper case, or first letter lower case, respectively.

Due to limitations of the parser, \1 and \x01 are the same which makes it difficult to use literal bytes with values lower than 10 in subst. Therefore \0 aways means a literal byte (incompatible change from earlier version!) and \1 through \9 mean literal bytes if they are larger than the number of sub-expressions. To get a literal & or \ or / in the substitution write \& or \\ or \/.

If width is specified, it limits the number of characters processed. If the - flag is used (i.e. width looks like a negative number) only the last width characters are processed, else the first. Without width (or 0) all available characters are processed.

If precision is specified, it indicates which matches to replace. With the + flag given, precision is the maximum number of matches to replace. Otherwise precision is the index (counting from 1) of the match to replace. Without precision (or 0), all matches are replaced.

When replacing multiple matches, the next match is searched directly after the currently replaced string, so that the subst string itself will never be modified recursively. However if an empty string is matched, searching advances by 1 character in order to avoid matching the same empty string again.

In input this converter pre-processes data received from the device before following converters read it. Converters preceding this one will read unmodified input. Thus place this converter before those whose input should be pre-processed.

In output it post-processes data already formatted by preceding converters before sending it to the device. Converters following this one will send their output unmodified. Thus place this converter after those whose output should be post-processed.

Examples:

%#+-10.2/ab/X/ replaces the string ab with X maximal 2 times in the last 10 characters. (abcabcabcabc becomes abcXcXcabc)
%#/\\/\// replaces all \ with / (\dir\file becomes /dir/file)
%#/..\B/&:/ inserts : after every second character which is not at the end of a word. (0b19353134 becomes 0b:19:35:31:34)
%#/:// removes all : characters. (0b:19:35:31:34 becomes 0b19353134)
%#/([^+-])*([+-])/\2\1/ moves a postfix sign to the front. (1.23- becomes -1.23)
%#-2/.*/\U0/ converts the previous 2 characters to upper case.

15. MantissaExponent DOUBLE converter (%m)

This exotic and experimental format matches numbers in the format [sign] mantissa sign exponent, e.g +123-4 meaning 123e-4 or 0.0123. Mantissa and exponent are decimal integers. The sign of the mantissa is optional. Compared to the standard %e format, this format does not contain the characters . and e.

Output formatting is ambigous (e.g. 123-4 versus 1230-5). I chose the following convention: Format precision defines number of digits in mantissa. No leading '0' in mantissa (except for 0.0 of course). Number of digits in exponent is at least 2. Format flags +, -, and space are supported in the usual way (always sign, left justified, space instead of + sign). Flags # and 0 are unsupported.

16. Timestamp DOUBLE converter (%T(timeformat))

This format reads or writes timestamps and converts them to a double number. The value represents the number of seconds since 1970 (the UNIX epoch). The precision of a double is large enough for microseconds (but not for nanoseconds). This format is probably used best in combination with a redirection to the TIME field. In this case, the value is converted to EPICS timestamps (seconds since 1990 and nanoseconds). The timestamp format understands the usual converters that the C function strftime() understands. In addition, fractions of a second can be specified and the time zone can be set in the format string.

Example: %(TIME)T(%d %b %Y %H:%M:%.3S %z) may print something like 3 Sep 2010 15:45:59 +0200.

Fractions of a second can be specified as %.nS (seconds with n fractional digits), as %0nf or %nf (n fractional digits) or as %N (nanoseconds). In input, n is the maximum number of digits parsed, there may be actually less digits in the input. If n is not specified (%.S or %f) it uses a default value of 6.

In input, the time zone can be specified in the format like %+hhmm or %-hhmm for cases where the parsed time stamp does not specify the time zone, where hhmm is a 4 digit number specifying the offset in hours and minutes.

In output, the system function strftime() is used to format the time. There may be differences in the implementation between operating systems.

In input, StreamDevice uses its own implementation because many systems are missing the strptime() function and additional formats are supported.

Day of the week can be parsed but is ignored because the information is redundant when used together with day, month and year and more or less useless otherwise. No check is done for consistency.

Because of the complexity of the problem, locales are not supported. Thus, only the English month names can be used (week day names are ignored anyway).